Classifier-Free Guidance

Also known as CFG

Classifier-free guidance is a way to make diffusion and flow-matching models follow conditions such as text prompts without using a separate classifier at sampling time. Work adjusts guidance strength across timesteps, spatial regions, or tokens, and explores alternatives such as informative negative conditions, geometric controls, and training-time alignment of conditional and unconditional branches. Methods are tested in settings including text-to-image composition, image editing, video generation, speech, and conditional image benchmarks. The central challenge is balancing condition alignment against fidelity, diversity, and structural quality: stronger guidance can help follow prompts but can also cause oversaturation or reduce diversity, while newer methods do not improve consistently across models and evaluations.

Latest papers

67 papers10 papers in the last four weeks, down 29% on the four weeks before. 0.1% of all new papers.

Guidance Is a Field, Not a Scalar: Interval-Integrated Classifier-Free Guidance for One-Step Video Generation
Nov 5, 2026Zijie Meng, Xixin CaoAdaptive CFG Classifier-Free Guidance
Peking University
Decompose the Distillation: Interpretable Single-Pass Guidance for Diffusion Models
Nov 3, 2026Wuyang Zhang, Xiaolin XuDiffusion Guidance Classifier-Free Guidance
Department of Electrical and Computer Engineering Northeastern University, Boston, MA, USA
Learning to Be Guided: Shaping Latent Geometry for Classifier-Free Guidance
Oct 16, 2026Haoming Liu, Junhao Zhu, Shaoyi Zheng +4Latent Space Geometry Diffusion-Based Image Generation
New York University · New York University Shanghai · Georgia Institute of Technology
Rethinking Classifier-Free Guidance for TI2I Flow Matching
Oct 15, 2026Andreas Floros, Elena Koritska, Teofilo de Campos +4Classifier-Free Guidance Text-Guided Image Editing
Imperial College London · Microsoft
Learning $ω$ -Powered Conditional Distributions with Importance-Weighted Flow Matching
Oct 12, 2026Maciej Falkiewicz, Frantzeska Lavda, Jason Ramapuram +1Conditional Distribution Estimation Conditional Flow Matching
Computer Science Department, University of Geneva · HES-SO/HEG Genéve · Google Deepmind (work done at Apple)
GuidanceNFT: NFT-Based Reward Fine-Tuning of Guided Diffusion Models
Oct 11, 2026Jiayin Liu, Zeyu Michael Li, Bingshuo Qian +1Diffusion Model Fine-Tuning T2I Diffusion Models
Tsinghua University · Duke University
Understanding Unified VLMs: From Attention Probing to Guidance Tuning for Multimodal Generation
Oct 10, 2026Zhengyang Luo, Xiaoyu Qiu, Zhiheng Tan +3Unified Multimodal Models Multimodal Generation
University of Michigan, Ann Arbor · The Ohio State University · Indiana University Indianapolis
TrajCFG: Token-Wise Classifier-Free Guidance via Denoising Trajectory Feedback
Oct 9, 2026Jingyu Liu, Hao Jiang, Yikang Ding +6Classifier-Free Guidance Diffusion Sampling
Renmin University of China · Kling Team, Kuaishou Technology
Opening the Black Box of Classifier-Free Guidance via Information Bottleneck
Oct 9, 2026Jiayang Gao, Tianyi Zheng, Jiayang Zou +5Adaptive CFG Classifier-Free Guidance
Shanghai Jiao Tong University, Shanghai, China · vivo BlueImage Lab, vivo Mobile Communication Co., Ltd., China · Stanford University, California, USA +1
VideoAWR: Reinforcing Video Flow Models with Decoupled-Timestep Advantage-Weighted Rectification
Oct 9, 2026Zheming Fu, Wenbin Ning, Ruizhe He +6RL for Video Generation Flow-Based Generative Modeling
Beihang University · Zhongguancun Academy · Joy Future Academy +2
Global Transport Couplings for Classifier-Free Guided Flows
Oct 6, 2026Katarina Petrović, Zander W. Blasingame, Danyal Rehman +5Classifier-Free Guidance Conditional Image Generation
AITHYRA · University of Oxford · Mila - Québec AI Institute +7
CACFG: Curvature-Aware Classifier-Free Guidance and Optimal Control
Oct 5, 2026Max Collins, Dasith de Silva Edirimuni, Jordan Vice +2Diffusion Model Guidance Classifier-Free Guidance
School of Physics, Maths and Computing The University of Western Australia Perth, WA 6009
Correcting Guided Diffusion Trajectories with Spectral Alignment
Oct 2, 2026Gihoon Kim, Taesup KimDiffusion Model Guidance Diffusion Sampling
Graduate School of Data Science, Seoul National University
Joint Branch-Space Transform Coding for Diffusion Activation Quantization with Classifier-Free Guidance
Oct 1, 2026Mingrun Jiang, Yuejia Liu, Zishan Shao +11Classifier-Free Guidance Diffusion Model Quantization
Duke University · Carnegie Mellon University · University of Florida +2
GFD-OPD: Guidance-Folded On-Policy Distillation of Diffusion Models Across Scales
Sep 30, 2026Zhenxing Zhang, Jiayan Teng, Wenxu Wu +6Diffusion Model Guidance Classifier-Free Guidance
Hefei University of Technology · Zhipu AI · Tsinghua University
ReGain: Restoring Subject Fidelity in Personalization on Synthetic Images
Sep 30, 2026Shubhang Bhatnagar, Ishan Bhatnagar, Viraj Shah +1Diffusion Model Guidance Classifier-Free Guidance
University of Illinois Urbana-Champaign
RA-CFGCache: From Branch-Level Criteria to Guided-Risk Control under Classifier-Free Guidance
Sep 29, 2026Yiming Liu, Ben Wan, Tongxuan Liu +5Classifier-Free Guidance Diffusion Model Caching
School of Software, Tsinghua University, Beijing, China · BNRist, Tsinghua University, Beijing, China · JD.com, Beijing, China
Attention-Scoped Guidance: Training-Free Spatial Control for Image Editing
Sep 28, 2026Zeyan Li, Wei Zhou, Hadi Amirpour +3Classifier-Free Guidance Text-Guided Image Editing
Shanghai Jiao Tong University · Cardiff University · University of Klagenfurt +1
SAGE: Subspace Alignment for Classifier-Free Guidance in Mixture-of-Experts Diffusion Models
Sep 28, 2026Boyu Zhang, Yangming Cheng, Ning Zhang +5Classifier-Free Guidance Diffusion Transformer
Alibaba Token Hub, Alibaba Group
DecFlowEdit: Self-Localized Flow-based Image Editing via Guidance Decoupling
Sep 28, 2026Zheyuan Zhan, Can Wang, Jiawei Chen +4Flow Matching Classifier-Free Guidance
Zhejiang University · University at Buffalo · University of California, Berkeley
Asymmetric Classifier-Free Guidance for Target-Speaker ASR
Sep 24, 2026Yiwen Guan, Jacob WhitehillAutomatic Speech Recognition Classifier-Free Guidance
Worcester Polytechnic Institute, Massachusetts, USA
Learnable Classifier-Free Guidance Null Embeddings for Enhanced Controllable Speech Synthesis
Sep 21, 2026Biel Tura Vecino, Yoach Lacombe, Julian Weber +4Classifier-Free Guidance TTS Synthesis
Cantina Labs
Classifier-Free Guidance in Flow Matching: Non-Autonomous Potentials, Overshoot, and Posterior-Mean Control
Sep 21, 2026Jishen Peng, Zheng MaFlow Matching Classifier-Free Guidance
School of Mathematical Sciences, Shanghai Jiao Tong University, Shanghai, China · Institute of Natural Sciences, MOE-LSC, Shanghai Jiao Tong University, Shanghai, China · Qing Yuan Research Institute, Shanghai Jiao Tong University, Shanghai, China +1
Geometry-Aware Diffusion Guidance via Curvature-Adaptive Tubular Correction
Sep 18, 2026Enze Jiang, Jinwei He, Zheng MaDiffusion Guidance Diffusion-Based Inverse Problems
School of Mathematical Sciences Shanghai Jiao Tong University Shanghai, China
Efficient Fairness Auditing Across Guidance Scales in Text-to-Image Diffusion Models via Causal Abstraction
Sep 8, 2026Nabila Tasfiha Rahman, Rajatsubhra Chakraborty, Depeng Xu +1Classifier-Free Guidance Algorithmic Fairness
University of Arkansas Fayetteville, AR, USA · University of North Carolina at Charlotte Charlotte, NC, USA
Particle Dynamics of Flow Matching and Classifier-Free Guidance from a Stagewise Geometry Perspective
Sep 7, 2026Jian-Feng Cai, Zhengyi Su, Chao WangFlow Matching Classifier-Free Guidance
Department of Mathematics The Hong Kong University of Science and Technology Hong Kong SAR, China · Department of Statistics and Data Science Southern University of Science and Technology Shenzhen, Guangdong, China
PathGuide: Dynamic Classifier-Free Guidance via On-Policy Transport Alignment
Aug 29, 2026Avishag Nevo, Tamir HazanAdaptive CFG Inference-Time Guidance
Technion - Israel Institute of Technology
Unlocking Multimodal Protein Language Models at Inference Time
Aug 26, 2026Yi Zhou, Qipeng Wang, Yunqing Liu +3Inference-Time Search Protein Language Models
The Hong Kong Polytechnic University · The Hong Kong University of Science and Technology (Guangzhou)
DCGC: Draft-Conditioned Global Correction for Complex Reasoning with Masked Diffusion Models
Aug 26, 2026Minhae Oh, Nakyung Lee, Jungwoo LeeLLM Self-Correction Diffusion Language Model Inference
Department of Electrical and Computer Engineering, Seoul National University
When Guidance Goes Off-Scale: Recalibrating Diffusion Transformers under Analog Compute-in-Memory Nonidealities
Aug 20, 2026Wenshuai Yao, Wenyong ZhouDiffusion Transformer Diffusion-Based Image Generation
School of Integrated Circuits, Peking University, Beijing, China · Department of Electrical and Computer Engineering, The University of Hong Kong, Hong Kong SAR, China

Nov 5, 2026J/K move · Enter open · S save

Guidance Is a Field, Not a Scalar: Interval-Integrated Classifier-Free Guidance for One-Step Video Generation

Zijie Meng, Xixin Cao

Peking University

Adaptive CFG Classifier-Free Guidance One-Step Generative Modeling Single-Step Video Generation MeanFlow

Open paper AskPDF arXiv

One-step generators built on MeanFlow synthesize a video in a single function evaluation, yet inherit a rigid treatment of classifier-free guidance (CFG): the guidance scale is frozen during training or, at best, exposed as a single scalar. This conflicts with mounting evidence that the optimal guidance is not a constant but a field that varies with the noise level, with spatial content, and, for video, with frame position. Every existing remedy presupposes a multi-step sampler along which guidance can be re-scheduled--an assumption one-step models violate by construction. We resolve this tension through a single observation: because the MeanFlow target is an integral over a noise interval, an arbitrary guidance field can be absorbed into the target itself and compiled into one evaluation. We formalize this as the interval-integrated CFG identity, prove that the compiled guidance carries no schedule-discretization error and can thus match or exceed any finite multi-step sampler, and instantiate it as AGAVE, which equips the average-velocity field with a learnable temporal schedule, a free motion-saliency spatial modulation, and a frame-position asymmetry. Adding fewer than ten scalars at zero inference cost, AGAVE surpasses all one-step baselines on VBench, UCF-101 and MSR-VTT while matching multi-step guided sampling. Project repository: https://github.com/YMlinfeng/AGAVE.